Clinical biomarker-based biological aging and risk of cancer in the UK Biobank

Abstract

Background: Despite a clear link between aging and cancer, there has been inconclusive evidence on how biological age (BA) may be associated with cancer incidence.

Methods: We studied 308,156 UK Biobank participants with no history of cancer at enrolment. Using 18 age-associated clinical biomarkers, we computed three BA measures (Klemera-Doubal method [KDM], PhenoAge, homeostatic dysregulation [HD]) and assessed their associations with incidence of any cancer and five common cancers (breast, prostate, lung, colorectal, and melanoma) using Cox proportional-hazards models.

Results: A total of 35,426 incident cancers were documented during a median follow-up of 10.9 years. Adjusting for common cancer risk factors, 1-standard deviation (SD) increment in the age-adjusted KDM (hazard ratio = 1.04, 95% confidence interval = 1.03-1.05), age-adjusted PhenoAge (1.09, 1.07-1.10), and HD (1.02, 1.01-1.03) was significantly associated with a higher risk of any cancer. All BA measures were also associated with increased risks of lung and colorectal cancers, but only PhenoAge was associated with breast cancer risk. Furthermore, we observed an inverse association between BA measures and prostate cancer, although it was attenuated after removing glycated hemoglobin and serum glucose from the BA algorithms.

Conclusions: Advanced BA quantified by clinical biomarkers is associated with increased risks of any cancer, lung cancer, and colorectal cancer.

Fig. 1. Correlations between biological age measures and chronological age in UK Biobank (n = 308,156).

The KDM residual and PhenoAge residual were computed by regressing out chronological age (as a natural spline term with three degrees of freedom) from the KDM-biological age and PhenoAge, respectively. HD homeostatic dysregulation, KDM Klemera-Doubal method.

Fig. 2. Time-varying hazard ratios for risk of cancers per 1 standard deviation increase in biological age measures in UK Biobank (n = 308,156).

a PhenoAge residual and risk of any cancer; b PhenoAge residual and risk of breast cancer in women; c KDM residual and risk of colorectal cancer; d PhenoAge residual and risk of colorectal cancer. Only the models with evidence of non-proportional hazards (P < 0.05) are shown. Estimates were obtained by including interaction terms between the exposure and the time variable (i.e., attained age, split into 5-year intervals) in the Cox models. The shaded areas indicate 95% confidence intervals (HRs outside the boundaries between 0.6 and 1.6 are not shown). The proportional hazards assumption of the biological age measures fitted in the Cox models was tested using Schoenfeld residuals. All models were adjusted for age (time scale), birth year, sex, baseline assessment center, ethnic background, body mass index, smoking status, physical activity level, alcohol consumption, education level, deprivation index quintiles, and the cancer-specific covariates as detailed in the footnote of Table 3 and in Supplementary Table 3. HR hazard ratio, KDM Klemera-Doubal method, SD standard deviation.

Fig. 3. Dose-response relationships between biological age measures and risk of cancer in UK Biobank (n = 308,156).

a Any cancer; b Prostate cancer in men; c Lung cancer; d Colorectal cancer. Only the models with evidence of non-linearity (P < 0.05) are shown. The black solid lines represent hazard ratios and the corresponding 95% confidence intervals (shaded areas) estimated using restricted cubic spline Cox regression models with knots at the 25th, 50th, and 75th percentiles. The dashed lines represent the estimates obtained from models assuming linear relationships. The mean of each BA measure was used as the reference value. P values for non-linearity were from likelihood ratio tests comparing the spline models with linear models. All models were adjusted for age (time scale), birth year, sex, baseline assessment center, ethnic background, body mass index, smoking status, physical activity level, alcohol consumption, education level, deprivation index quintiles, and the cancer-specific covariates as detailed in the footnote of Table 3 and in Supplementary Table 3. HD homeostatic dysregulation, KDM Klemera-Doubal method, SD standard deviation.